in addition to @Juan_Olano‘s great advice:
I am convinced it’s all about the right use case. Often when people see some first positive results on their use case, my experience is that suddenly plenty of potential ways to improve your solution pop up, be it: tuning the model or advancing the data or automatising the system etc.
I would rather spend a little more effort for carefully evaluating your first pilot project that you want to start - maybe it’s the Kaggle challenge that is of highest interest to you, but maybe you also want to search something very close to your job-specific challenges (e.g. from time series domain or computer vision or embedded AI on the edge trained centrally in the cloud…) where you see that this project is a strategic fit and which also might help you to keep your motivation high also for the months and years to come.
Actually some recipes that might be worth considering can be found here:
Hope that helps.
Best regards
Christian